AI-Driven Decision Platforms Live in the Moment
The emerging market for AI-driven digital decision-making is grounded in new applications developed using machine learning models designed to help analysts crunch data that would allow their bosses to make the right decision, whether its engaging with customers or managing a complex supply chain.
Based on the assumption that AI-based decision-making is probabilistic (a machine learning model might be 97-percent accurate, but what about the other 3 percent?) proponents of the “decision-first” approach argue companies need to stop wallowing in data, get the lead out and make timely, informed decisions.
“It all comes together on the decision” to take a specific action, said John Rymer, principal analyst with Forrester Research. Rymer and his research colleague Mike Gualtieri made their case for AI-based decision management during this week’s Tibco Software conference in Chicago.
The researchers argued there is a decision-making role for “pragmatic AI” that is narrow in scope but superior to humans in terms of speed. Its application promises to help companies emerge from “analysis paralysis” and focus identifying the most important decisions—then pull the trigger.
Emerging digital decision platforms are built on large volumes of enterprise data used to train machine learning models. Those models serve as the building blocks of AI-based applications. “Applications are where digital decisions live,” Gualtieri said
The market analysts are quick to acknowledge that the decision aids are far from full-proof, but the combination of predictive machine learning models and data scientists usually arrives at better decisions than does a human alone. For example, AI-based decision apps excel at predicting failures that ultimately influence decisions.
Vendors also have made the case for other approaches to achieve real-time decision-making as interactive and other schemes fail to keep pace with customers’ expectations. The Forrester researchers argue that agile “decision-first” platforms must be developed using machine learning models that represent the building blocks of AI apps.
“This is really application development and the developers have to be involved,” said Rymer.
For now, vendors are beginning to offer early development tools as the decision management ecosystem is fleshed out. It remains an emerging market, but the analysts see progress on development tools that would automate some platform components and generally make data science teams more efficient.
Ultimately, data remains a “huge prerequisite,” the researchers stressed, and enterprises have plenty of customer and other data to work with. That resource would allow platform developers to train machine- and deep-learning systems that would underlay “AI-infused applications.”
Meanwhile, data analysts must establish context for potential applications and, once a platform is scaled up, different use cases. “This is not just analytics for the sake of analytics,” Gualtieri stressed. The real-time capability could then allow the platform to arrive at a decision in real time—for example, while a demanding customer is browsing on a retail web site.